Physical layer cognitive engine for multi-antenna systems

In this paper, we propose a two-step cognitive engine that aims to learn its physical layer capabilities in various channel conditions and optimize the delivery of packets in a multi-antenna radio. The first step learns the capabilities of the available techniques, and the second step finds the configuration that best meets the goals of the radio, subject to the channel conditions. This approach allows a good segmentation of the problem and provides the freedom to research each part independently of the other. We propose a reference design based on Bayespsila rule and a brute force search for the learning and optimization stages, respectively. This reference design is to be used as a baseline for the comparison of future implementations. In this paper we also investigate the applicability of the Naive and Semi-Naive Bayespsila models for learning and a binary search for optimization. The Semi-Naive Bayespsila model and a binary search are found to provide good alternatives to the learning and optimization segments of the reference design by requiring the estimation of fewer parameters and fewer searches, respectively. However, the non-reference system sacrifices performance optimality for speed and memory savings as compared to the reference design.